GEN_LINEAGE← Back to home
About Genlineage

EVERY GENERATION HAS A HISTORY

Genlineage is version control for generative media. Teams use it to build and manage AI generation pipelines for images, video, audio, and music. Every generated asset is versioned, evaluated, connected to its lineage, and cryptographically signed. Think of it as Git for AI generated media.

Why it exists

GENERATIVE PIPELINES FORGET

A modern campaign touches dozens of generations across multiple models: storyboard frames, animated shots, voiceover takes, music beds, final cuts. The prompts that produced them live in chat histories. The parameters are gone. Which take was approved? Which model made it? What did the failed attempts cost?

Genlineage answers those questions structurally. Every generation is a commit with a complete recipe. Every remix is an edge in a graph. Every merge passes a quality gate, and every merged asset carries a signed, independently verifiable record of where it came from.

What it is

FOUR IDEAS, ONE LEDGER

01

Media commits

Every generated asset is recorded as an immutable commit: the bytes, plus the complete recipe that produced them model, prompt, parameters, seed, parents, and evaluation scores. The asset's own sha256 hash is its permanent address.

02

The remix DAG

Commits reference their parents, so every remix, variation, and re-cut becomes an edge in a directed graph. Any asset traces back to its origin in one walk and branching a campaign only regenerates what actually changed.

03

The quality gate

An evaluator model reviews every output against its spec before it merges. Below threshold, the critique is folded into revised parameters and retried; if a generation path keeps failing, the pipeline fails over automatically.

04

Content credentials

A signed manifest ships beside every merged asset its recipe, its lineage, and an ed25519 signature. Anyone can verify where an asset came from without trusting our word for it.

How a run works

FROM BRIEF TO SIGNED ASSET

You give the studio a brief. A planning model decomposes it into typed steps frames, shots, voiceover, music, final cut each with its own spec. Every step is generated, scored by an evaluator against its spec, retried or failed over if it misses the bar, then hashed, signed and committed to the ledger. The result is not just a folder of outputs: it's a graph you can branch, audit, and verify.

01
Plan
02
Generate
03
Evaluate
04
Retry / failover
05
Commit + sign
What we believe

PRINCIPLES

Provenance is not a feature

It's the substrate. In a world of infinite generation, the history of an asset is what makes it trustworthy, reusable, and worth paying for.

Never pay twice

Content-addressing means identical work is stored once and referenced everywhere. Remix a campaign and unchanged steps cost nothing.

Fail clearly, recover quietly

Every rejected generation is kept and logged. Retries and failovers happen automatically but the full record of what was tried is always one click away.

Verify, don't trust

Signatures and append-only logs mean our own database is not the source of truth. The ledger is.

See it yourself

START TRACING

Create an account →View pricing